SettingsSign in

Glossary

Plain definitions for AI terms you'll see across tools, docs, and this directory.

A

AI agent: A system that plans multi-step tasks, calls tools, and iterates until a goal is met — not just a single chat reply.

C

Context window: How much text (in tokens) a model can consider in one request — including your prompt, files, and its reply.

E

Embedding: A numeric vector representing meaning. Similar texts have similar vectors — used for search and clustering.

F

Few-shot prompting: Including a few input→output examples in the prompt to teach format and behavior without fine-tuning.
Fine-tuning: Additional training on a smaller dataset to specialize a base model for a domain, tone, or task.

G

Grounding: Tying answers to verified sources — web search, your files, or tool output — instead of memory alone.

H

Hallucination: When a model states false information confidently. Mitigate with citations, RAG, and verification steps.

L

LLM (Large Language Model): A neural network trained on vast text to predict the next token. Powers chat assistants, coding tools, and many AI products.

M

MCP (Model Context Protocol): An open standard for connecting AI clients to external tools and data via MCP servers (filesystem, GitHub, databases, etc.).
Modality: Input/output types a model handles: text, code, image, voice, video. Multimodal models support several.

P

Prompt engineering: Crafting instructions, examples, and structure so models produce reliable, useful outputs.

R

RAG (Retrieval-Augmented Generation): Fetch relevant documents first, then ask the model to answer using those sources — reduces hallucinations on private data.

S

System prompt: Hidden instructions that set persona, rules, and constraints for every turn in a conversation.

T

Token: The unit models read and write — roughly ¾ of an English word. Pricing and context limits are often measured in tokens.
Tool use / function calling: When a model returns structured calls (e.g. search, run code) that the host app executes and feeds back as results.

V

Vector database: Storage optimized for similarity search over embeddings — common backend for RAG and semantic search.